This is a naïve (unoptimized) implementation of pix2pix in Unity.

This implementation supports the weight data format used in Christopher Hesse's interactive demo. Pick one of the pre-trained models, or you can train your own model with using pix2pix-tensorflow.

At the moment, it's implemented in naïve C# and hasn't been optimized by any mean. It takes about 5 minutes to inference a single image. Personally, I'm optimistic about performance because there are many established approaches to optimize neural network inference.